dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Oct 21, 2011 at 11:27:57AM +0200, Anders Logg wrote:
> On Fri, Oct 21, 2011 at 10:41:57AM +0200, Martin Sandve Alnæs wrote:
> > If this cannot be fixed quickly, can you please revert the commit that
> > caused it?
> >
> > In general, quickly fixing or reverting when stuff is broken is the
> > only way we can keep a consistently green buildbot. Having the
> > buildbots in a red state over time hinders further development in
> > maintrunk and/or hides problems caused by additional commits. This
> > time the build problem did hide test breakage caused by both me and
> > Anders, and the fixes were then much delayed. I think it is worth
> > considering requiring quick revert of all developers (I know I've
> > sinned there myself).
>
> Sorry for breaking the buildbot, but I've had trouble with my own
> personal buildbot not running at all so I couldn't test there first.
> (And the thing that currently breaks works fine here with me.)
>
> I'm not sure reverting now is a good idea since it is a big change
> that is important for the book (cleanup of the Krylov methods and
> preconditioner options).
>
> It looks to me like something needs to be updated or rebuilt on the
> buildbot. The unit test that fails on the buildbot (quadrature on cut
> cells using CGAL) works fine here, and also for Andre. Maybe the
> buildbot just needs a new installation of CGAL and/or boost?

The error message seems to indicate a problem with boost::assign in
combination with CGAL. Maybe somehow the inclusion of

  boost/assign/list_of.hpp

in uBLASFactory.h and MTL4Factory.h propagated to the unit test and
that breaks with CGAL.

I've worked around the inclusion of this header file now (only
included in .cpp files). Can those of you for whom this breaks try
again?

--
Anders